Examples of unusual ionospheric observations by the CSES prior to earthquakes

The CSES (China seismic electromagnetic satellite) was launched on February 2, 2018 in a circular polar orbit at an altitude of ~507 km. One of the main objectives of CSES is to search for and characterize ionospheric perturbations that can be associated with seismic activities, to better understand the generation mechanism of such perturbations. Its scientific payload can measure a broad frequency range of electromagnetic waves and some important plasma parameters. This paper is a first‐hand study of unusual observations recorded by the CSES over seismic regions prior to four earthquakes with M >7.0 since the satellite's launch. CSES detectors measured irregularities near the epicenter of these four earthquakes. It is already clear that data from instruments onboard the CSES will be of significant help in studies of characteristics of ionospheric perturbations related to earthquakes and their generation mechanisms.


Introduction
Since the l980s, many countries have developed satellites capable of making electromagnetic observations. Spatial electromagnetic precursors to seismic events have been reported since then (Larkina et a1., 1983;Chmyrev et a1., 1989;Parrot and Mogilevsky, 1989;Molchanov et al., 1993), serving as a new observing technique for studying seismic electromagnetic effects (Zhang XM et al., 2009a). Many studies have focused on seismo-electromagnetic effects and disturbances in the ionosphere, and on possible generation mechanisms. The DEMETER satellite was the first satellite for earthquake monitoring from space (Parrot, 2006). It has provided data that has advanced research in this field.
Many ionospheric electromagnetic, plasma, and energy particle perturbations related to seismic activities have been detected. For the electromagnetic field, data in the range of ELF, VLF, and ULF frequencies have been studied. Zhang XM et al. (2009b) and Bhattacharya et al. (2007) found electric and magnetic field variations in the ULF/ELF range (40 Hz,110 Hz,350 Hz,and 400 Hz). Spectrogram results clearly indicate strong emissions before earthquakes. This is in agreement with the work done by Larkina et al. (1983), Bhattacharya and Gwal (2005) and Parrot et al. (2005), who earlier detected electromagnetic emissions over earthquake regions. Zeng ZC et al. (2009) and Zhu T and Wang LW (2011) studied average power spectrum density (APSD) of the electric field. In most cases, APSD in the range of ULF/ELF increased, while APSD in the range of VLF decreased. Zhu T and Wang LW (2011) also found that the amplitudes increased substantially before the occurrence of an earthquake. It then decreased and dropped almost to their minimum when the earthquake occurred. Akhoondzadeh et al. (2010) studied the cut-off frequency of the electromagnetic field. In their study, perturbations are observed near the epicenter of the quake in the form of change or transition in cut-off frequency at around 350-400 Hz range in ELF. The electric field is more prominent than the magnetic field spectrogram. For plasma observation data, the enhancement of both ionospheric plasma density and temperature are all found before the occurrence of earthquakes. Ion temperature, ion density, and electron density show uniform characteristics (Zhang XM et al., 2009c;Ouyang XY et al., 2011;Liu J, 2013;Liu J et al., 2014). GPS-TEC and f o F 2 measurements have also been instrumental to improving the understanding of seismo-ionospheric variations, as demonstrated by Liu JY et al. (2002, Lognonné et al. (2006) and Ding ZH et al. (2010). Using IDP (an energetic particle detector) data from the French DEMETER satellite, Huang JP et al. (2010) analyzed energetic particles around the time of the M8.8 Chili earthquake and found increased night-time fluxes prior to the earthquake.
Statistical analyses have been used to study ionospheric parameters. Afonin et al. (1999) correlated ionospheric electron density perturbations and earthquakes, using data obtained by satellite Intercosmos 24 (Afonin et al., 1999). A statistical study of electric field variations up to 10 kHz (Němec et al., 2008(Němec et al., , 2009) reported a statistically significant decrease of wave intensity several hours before the occurrence of an earthquake. Píša et al. (2012Píša et al. ( , 2013 arrived at the same conclusion using the complete DEMETER data set (9000 earthquakes) and found wave intensity decreases to a frequency of nearly 1.7 kHz during the nighttime. Zeren ZM et al. (2012) conducted statistical analysis of the characteristics of ELF/VLF disturbances of the magnetic field when M s is greater than or equal to 7.0, based on northern hemisphere earthquakes occurring between 2005 and 2009. The authors found that for 42% of these earthquakes, perturbation amplitudes increased 30 days before the occurrence of an earthquake. For 35% of these earthquakes, however, decreased perturbation was observed. For the remaining 23%, no significant difference in perturbation was observed prior to the earthquake. Using a superposed epoch method, He Y et al. (2011) demonstrated that there is an increase of electron density close to the epicenter during night time. Statistical analyses have been performed using ion density peaks in the DEMETER data set (6.5 years) (Parrot, 2011(Parrot, , 2012Parrot, 2012, 2013). The authors have shown that the number of good fits increases with the earthquake magnitude. In addition, it is found that the number of perturbations is higher for the day when earthquakes occur. Perturbations then slowly decrease after the earthquake. Yan R et al. (2017) found anomalous ionospheric perturbations related to earthquakes of magnitude larger than 5. The perturbations generally appear within a range of 200 km from the epicenter and mainly 5 days before the earthquakes. Using more than 6 years data observed in DEMETER, Zhang X et al. (2013) found that there are increases in the number of electron bursts prior to the events; during the entire operation period of the satellite, electron burst precipitation occurred before each strong earthquake with magnitude over 7.0. Liu JY et al. (2009) examined earthquakes during the 10-year period from 1 May 1998 to 30 April 2008 and performed a statistical analysis of GIM TEC and found that the GIM TEC above the epicenter decreased 3-5 days prior to the occurrence of 17 earthquakes with M≥ 6.3.
Both positive and negative ionospheric effects of these parameters are observed. According to the Lithosphere-Atmosphere-Ionosphere Coupling (LAIC) models, these effects can be explained by anomalous electric fields on the ground surface in either the positive or the negative direction (Pulinets, 2009). Analyses of ionospheric perturbations are provided by Hayakawa et al. (2004), Freund et al. (2009), Freund andFeeund (2015), Freund (2011), Pulinets et al. (2015, and De Santis et al. (2015). Ionospheric density variations could be triggered by changes of the electric field and currents in the global electric circuit between the bottom of the ionosphere and the Earth's surface, where electric charges occur, associated with stressed rocks (Kuo et al., 2011;Kelley et al., 2017). Other hypotheses and modeling can be found in monographs (Sorokin et al., 2015;Hayakawa, 2015), and references therein.
The China Seismo-Electromagnetic Satellite (CSES), which is also named as ZHANGHENG-1, was successfully launched on February 2, 2018 (Shen XH et al., 2018). The CSES is the first Chinese spacebased platform for both earthquake observation and geophysical field measurement. It is the first satellite of the planned Chinese space-based geophysical field observation system and is expected to have a wide application in earthquake science, geophysics, space sciences, etc. The CSES can measure various physical parameters including electromagnetic field, electromagnetic waves, ionospheric plasma parameters, and high energy particles. One of its scientific objectives is to monitor and study ionospheric perturbations that appear to be associated with seismic activities, especially events of devastating size. The CSES is currently in its orbit-test phase. After the orbit-test, data obtained from the CSES will be published for public access (www.leos.ac.cn).
Although statistical analysis can help identify ionospheric changes associated with subsequent occurrence of earthquakes, studying individual earthquake events is important to the development of scientific understanding of these ionospheric variations. During the first few months of CSES operations, several earthquakes with M>7.0 were observed. The aim of this study is to present examples of ionospheric perturbations recorded by the various experiments carried by CSES when it was flying over active seismic regions during the orbit-test period. The scientific payload of the satellite and data recorded by the payload's various components are briefly described in Section 2. Analysis of data collected before, during, and after earthquakes is presented in Section 3, followed by discussion and conclusions in Section 4.

The CSES Satellite and Data Production
The CSES is a sun-synchronous satellite orbiting at a height of approximately 507 km with a descending node of 14:00 local time. The scientific payload of the CSES is composed of several instruments that provide a nearly continuous survey of ionospheric plasma, waves, and energetic particles. The EFD, an electric field detector, uses four electric probes to measure the three components of the electric field in a frequency range from DC up to 3.5 MHz. SCM, a search-coil magnetometer, measures the three components of the magnetic field in a frequency range from 10 Hz up to 20 kHz. The HPM, a high precision magnetometer, measures the vector of the Earth's magnetic field with a bandwidth from DC to 15 Hz. The LAP, a Langmuir probe, allows access to the electron density and temperature. The PAP, plasma analyzer package, measures ion density, composition, temperature, and flow velocity. The HEPP and Italian HEPD, high energetic particle package and detector, measure the proton energy spectrum, electron energy spectrum, and pitch angle. The GNSS-RO, GNSS occupation receiver, measures TEC, electron density, and atmosphere density. The TBB, tri-band beacon, measures TEC and electron density. Details about these instruments can be found in (Ambrosi et al., 2018;Cao JB et al., 2018;Cheng BJ et al., 2018;Liu C et al., 2018;Huang JP et al.,2018;Chen L et al., 2018;Chu W et al.,2018;Yan R et al., 2018;Cheng Y et al., 2018 ).
During its five years' operation time, the CSES will acquire many kinds of data, e.g., multi-band waveform, spectrum of electric andmagnetic fields, in-situ plasma parameters (including densities and temperatures of electron and ion), electron density profiles and tomography, and energetic particle flux and energy spectra. There are five levels of data, which are described below: Level-0: The raw data of payloads generated after a series of processings including frame synchronization, de-randomization, decoding, and de-formatting. All redundant data are removed in this level.

Level-1:
The data obtained after general error rejection, format conversion, and calibration of Level-0 data.

Level-2:
Physical quantity data with satellite orbit information after coordination system transformation and necessary data inversion of Level-1 data.
Level-3: Time sequential data in the frame of satellite orbits, gen-erated after resampling and spectral analysis based on Level-2 data.
Level-4: Global or regional dynamic observation data retrieved from Level-2 and Level-3 data, in terms of variation between recursive orbits and disturbances observed according to comparison with the background field.
Experimenters and guest investigators can have access to the database and download data by application to their different au-thorities.

Data Analysis and Results
During August 2018, four earthquakes occurred with M>7.0 (Table 1), according to the report from China Earthquake Networks Center (http://www.ceic.ac.cn/speedsearch). We took these four earthquakes as examples. Based on their depths, these four earthquakes can be classified as shallow, intermediate depth, and deep. Their positions are shown in Figure 1.
Many geomagnetic disturbances, such as magnetic storms, could significantly disturb the ionosphere. In general, Kp index and Dst index can be used to determine whether a disturbance is caused by a space event or an earthquake. If the Kp index is less than or equal to 3, or the Dst index is greater than or equal to -20 nT, the period is considered to be quite geomagnetically calm. To separate seismo-ionospheric perturbations from geomagnetic fluctuations, Dst and Kp index are monitored. These indices are widely regarded as measures of solar wind-magnetosphere interaction and are used in examining interplanetary and solar phenomena . Kp index and Dst index data related to solar activities and geomagnetic conditions during August, 2018, are shown in Figure 2. Evidence of a severe magnetic storm from 25 August to 28 August is apparent.
Observation data from the satellite's EFD, SCM, LAP, PAP and HEPP associated with these four earthquakes are studied and dis-cussed below.

Case-I: Shallow Depth Earthquake
The first seismic event discussed in this paper is the Hunter earthquake, a shallow event that occurred about 300 km to the east of the island of New Caledonia in the southwest Pacific Ocean on August 29, 2018 at 03:51:54 UT. The magnitude of this earthquake is 7.1 and the depth is 20 km. The epicenter is located at geographic 21.95°S and 170.10°E. The revisited orbits close (less than 1000 km) to the epicenter are shown in Figure 3, in which the star indicates the epicenter of the quake. Panel (a) of Figure 3 corresponds to ascending orbits (nighttime), and the right panel (b) is for descending orbits (daytime). The blue curves represent footprints of the revisited orbits; the numbers near these curves indicate orbit number and time (date). For instance, in the number combination "31371-0827", '3137' is the orbit number, '1' is the sub-orbit (in this case, ascending orbit); '0827' means 27th August.  We used space difference methods (Liu, 2013) to analyze the nighttime electron density (N e ) observed by the LAP. Taking the epicenter as center, a ±15° area was divided into grids of 4° (longitude) × 2° (latitude). The background average (Ab) and standard deviation (σ) in each grid were then obtained using data collected in July. From 12 days prior to EQs to 2 days after EQs, averaged observation data (Ao) for each orbit inside each grid are obtained. The normalization can be calculated from the ratio of (Ao-Ab) and σ in each grid. Space difference results are shown in Figure 4, in which ±10σ (standard deviation) was selected as the threshold for abnormity. The nighttime N e space differences observed during August 17-31 are shown. X-axis and Y-axis are longitude and latitude respectively. The threshold of the difference is color-coded according to the scale on the bottom. The star represents the epicenter. Panels (a), (b), and (c) correspond to the following time intervals: August 17-21, 22-26, and 27-31.
However, a severe magnetic storm began on August 25 and lasted until August 27. The Dst and Kp indexes are large during these days. So it is difficult to distinguish whether the perturbation is caused by the magnetic storm or the earthquake. We found an intense enhancement on August 20 and an enhancement on August 17 (arrow in Figure 4a). These two increases appear before the earthquake and are near to its epicenter. Although the date of August 17, Kp index is equal to 4 during 03:00-06:00 UT, the date time passing by epicenter is around 15:00-15:30 UT where Kp is equal to 2. Hence the enhancement occurred on the 17th and 20th August is probably related to the Hunter Earthquake. The increases in Figure 4b that occurred on the 25th and 26th August are probably related to the magnetic storm. There is a decrease on August 27 (Figure 4c), two days before earthquake. Some abnormal increases are also found in the proton spectrum observed by the HEPP prior to the earthquakes. Data of all revised orbits near the epicenter (less than 10°) are compared and a significant increase is found. As shown in Figure 5, compared to data from other revisited orbits, on August 18 a significant electron flux increase is observed at almost all telescopes (it is not significant in the data from telescopes 1 and 9). Results of data from telescopes 4 and 5 are shown in Figure 5, in which the X-axis is latitude and the Y-axis is electron flux. Lines in different colors represent different orbits, orbits 29241, 30001, 30761, and 31521 ( Figure 3a). The Kp index on August 18 is at its minimum and the position is in the middle latitude, which is not affected by the aurora zone.
Anomalies in the magnetic field in some frequency ranges were also observed. Based on previous research (Zhang XM et al. 2009b;Bhattacharya et al., 2007;Zeng ZC et al., 2009;Zhu T and Wang LW et al., 2011), most abnormal magnetic field seismic signals observed by satellite are in a frequency range of DC -1000 Hz. The power spectrum density (PSD) in this frequency range has been, therefore, further studied. Similar to the space difference method, nighttime revisited orbits which are less than 30° from epicenter are identified for further analyses. Data from 10 days to 40 days before the occurrence of the earthquake are used as background. The ±30° area was divided into grids of 5° (longitude) × 2.5° (latitude). Average background average (Ab) in each grid was then obtained. The average values of observe data (Ao) in every 5 days using data 15 days prior to the EQs are also calculated. The difference between Ao and Ab is shown in Figure 6.  Figure 6c is observed, which is most probably caused by the magnetic storm. In addition, the revisited orbits that passed by the epicenter are during 14:00-15:30 UT. The Kp index values larger than 3 occur at 13:00-15:00 on August 15, at 4:00-6:00 in August 17, and at 19:00-21:00 in August 20. Considering these times and positions when the Kp index is large than 3, the increases observed to the southwest of the epicenter (arrow in Figure 6a) are probably due to the earthquake. The increases occurred on August 17 and 18.
The spectrum of the electric field in the range of DC-2.5 kHz was analyzed. First we calculate accumulation of power spectrum density (PSD) in each nighttime orbit within 10° of the epicenter. The time sequence of these accumulation values during August 10-28 along the frequency spectrum are shown in Figure 7, in which the X-axis is date and the Y-axis is frequency. The PSD accumulation value is color-coded according to the scale on the right. The panels from top to bottom correspond to the E x , E y , and E z components of electric field. It can be seen that there is a strong Ionospheric perturbations of N e within 10 days before and 2 days after this earthquake were studied, using the moving average method. The nighttime data collected on one orbit within ±10°1 0029241_2018-08-13 0030001_2018-08-18 0030761_2018-08-23 0031521_2018-08-28 0029241_2018-08-13 0030001_2018-08-18 0030761_2018-08-23 0031521_2018-08-28 around the epicenter were resampled by 0.5° in the direction along the latitude. The background average value (Ab) and standard deviation (σ) were calculated using data from the previous 10 orbits (about 10 days), taking Ab±2σ as the upper and lower thresholds. The average value of current resampled orbit data (Ao) is also shown at the same time (upper panel Figure 8). The relat- ive change (Ao-Ab)/σ is calculated when observation value is greater than the upper threshold or smaller than the lower threshold. If observation values are between the upper and lower thresholds, the relative change is then 0 (bottom panel Figure 8).
The upper panel of Figure 8 shows the variations in observed N e , background, and upper and lower thresholds (from August 11 to August 22), including during the earthquake (August 21). Increases on August 12 and August 16 are obvious. The variation for August 16 is more than 2σ. The Kp index during these revisited orbits concentrates on the time interval of 05:00-07:00 and is consistently less than or equal to 3. Hence, the two N e increases observed 5 days (August 16) and 9 days (August 12) before the earthquake are not due to magnetic storms. The increases observed on August 11 and August 20 were single points, not enough to be considered anomalies.
Disturbances on August 16 were observed also by the PAP: signi-ficant increases of oxygen density (N O+ ) and ion longitudinal drift velocity (V x ). These two parameters showed rapid increases in the region of 100 km above the epicenter; observed values drop significantly when the satellite leaves the epicenter region ( Figure 9). Similar to Figure 6, Figure 10 displays differences in the PSDmeasured magnetic component (E z ) at 500 Hz during August 7-11 (Figure 10a), 12-16 (Figure 10b), and 17-21 ( Figure 10c). PSD differences are color coded according to the scale at the bottom. The star indicates the epicenter. As the earthquake is approaching, the PSD close to the epicenter is observed to increase. In particular, to the east of the epicenter, the PSD increased to its maximum 0-5 days before the earthquake. However, it is noted that the position of increase is around the position of the orbit 29791 (August 17), as shown in Figure 11. The Kp index is 4 at 03:00-06:00 on August 17. Orbit No. 29791 passed by the epicenter at approximately 05:30 UT. Therefore, it is reasonable to argue that such an in- crease is related to magnetic activities. The increase on August 15 (Figure 10b and Figure 11) and the one on August 20 (Figure 10c and Figure 11) are more likely related to the earthquake. These increments are all to the west of the epicenter (arrows in Figure 10).

Case-III: Deep Depth Earthquakes
The first deep seismic event observed is the earthquake close to 138 km WNW of Iberia, Peru, which occurred on August 24, 2018, at 09:04:05 UT. The magnitude of this earthquake was 7.1, the depth is 600 km and the epicenter was located at geographic 10.95°S and 70.75°W. Similarly, the data of revisited orbits close (less than 1000 km) to the epicenter are identified.
Similar to Figure 7, the electric field spectrum in the range of DC-2.5 kHz was analyzed. The time sequence of observed values accumulated along the spectrum during August 10-19 is shown in Figure 12. An anomalous increment of PSD below 15 kHz is obvious on August 18. The increase is found in all three components. The magnetic activity condition on this day was quiet.
The fourth case study is a deep earthquake of magnitude 8.1 that occurred on August 19, 2018 at 00:19:37 UT close to Ndoi Island, Fiji (18.08°S, 178.06°W). For the proton spectrum observed by HEPP, we first calculate the background spectrum average value using revisited orbit data collected between 21 July and 19 August near the epicenter (less than 10°). The spectrum 0-5 days before the earthquake is then calculated (Figure 13). In Figure 13, the X-axis is the energy and the Y-axis is the proton spectrum flux. The blue curve displays background data collected from 21 July to 19 August. The red curve displays observations from 15 to 19 August, before the earthquake, showing that, compared to the relatively calm background spectrum, the proton energy spectrum in these days immediately prior to the earthquake was significantly disturbed.
Accumulated power spectrum densities (PSD) of the electric field for each nighttime orbit within 10° around the epicenter are also analyzed ( Figure 14). An increment of PSD in the range of 10 kHz-20 kHz on 17 August (ellipse in Figure 14) is observed. The increase is shown in all three E x , E y , E z components. The Kp index is found to be minimum.

Discussion and Conclusions
This study presents ionospheric perturbations observed during four earthquakes. These perturbations were recorded by the CSES's EFD, SCM, LAP, PAP and HEPP instruments. These results demonstrate that unusual features in waves, plasma, and energetic particle fluxes can be recorded when the satellite flies over regions that include epicenters of future earthquakes. All phenomena related to earthquakes in this study were observed during nighttime orbits. The deeper the earthquake depth, the fewer were the anomalous phenomena observed. The greater were the magnitude of the seismic disturbance, the greater the observed electromagnetic disturbances.
Conclusions from this study can be summarized as follows: (1) The same parameters may not be reflected in each earthquake, and the observed anomalies may not be the same as in previous events.
(2) Perturbations of different parameters can occur at various locations and times.
(3) The time and location of disturbances before an earthquake appear to vary from one event to another.
(4) Perturbations are found for some parameters after earthquakes, but not for every parameter in each earthquake.
All the measured parameters display similar background variations in absence of seismic activity since the ionosphere is affected by a number of other factors. Kp and Dst can be used to examine whether perturbations are from earthquakes or magnetic storms, but they cannot be used to rule out other possibilities. This is one of the challenges in studying earthquake precursors.
Based on results presented in this study, we conclude that CSES data are responsive to earthquake events. As the CSES becomes fully operational, it will facilitate detailed study of the relationship between ionospheric perturbations and earthquakes. Additional case studies will further advance understanding and predicting earthquake activity.   VLF spectrum around the epicenter ±10°a long the ascending orbit Figure 14. The time sequence of PSD accumulation for each orbit before earthquake.